There are two ways to make new memories. First, you attach a new K-line to all the agents that were recently active. Second, you attach a new K-line to older K-lines that were active recently. (NB: Important point -- now we see that K-lines can be attached to other K-lines too -- perhaps it is not bad now to think of K-line as a special kind of agent.) This builds K-line memory trees.
"The kinds of mental states that this `hierarchical' type of memory produces will be based more on stereotypes and default assumptions than on actual perceptions. Specifically, you will tend to remember only what you recognized at the time. So something is lost -- but there's a gain in exchange. These `K-line memory trees' lose certain kinds of details, but they retain more traces of the origins of our ideas."
Another really neat idea -- Minsky is going beyond association between ideas to higher-order associations between associations. What precedent is there for this idea?)
Pearl Street | Society of Mind Home Page | Dawson Home Page |